GDPx

Research firm SemiAnalysis has introduced GDPx, a new benchmark that evaluates AI agents across 11 professional domains, ranging from chip design to cybersecurity. Anthropic's Opus 5.5 tops the leaderboard, but with a score of just 12%, highlighting how far AI agents remain from expert-level performance on demanding real-world tasks. The low ceiling and breadth of tested domains are drawing attention from AI practitioners discussing how it compares with existing benchmarks and what the results say about current agent capabilities.

SemiAnalysis has introduced GDPx, a new framework for measuring frontier AI model capabilities based on the most economically valuable domains rather than traditional academic benchmarks. The proposal aims to assess how well advanced models perform on tasks that map directly to real-world productivity and GDP contribution, sparking debate among AI researchers about how model progress should be evaluated and commercialised.